Quantum Efficiency

A key performance parameter for an astronomical CCD is QE. CCDs are capable of detecting photons over a wide range, as shown in Fig. 4. The figure depicts direct silicon sensitivity, although wavelength-converting materials, e.g., phosphors, are also used. Several factors control silicon sensitivity: (1) High energies (X-rays) have a long absorption length, and sensitivity depends on the thickness of silicon. (2) Short-wavelength ultraviolet photons are absorbed close to the silicon surface and need back side illumination. (3) Photons of wavelength >100 nm are efficiently absorbed in a back side CCD with antireflection coating. (4) Photons of wavelength >400 nm can be detected by a front side illumination; electrodes limit response. (5) Photons of wavelength >700 nm have a long absorption length so thicker silicon helps; sensitivity is determined mainly by the thickness of silicon and partly by operating temperature (band gap shifts). (6) Photons of wavelength >1100 nm are not detected because of the silicon band gap limit, which also has a temperature dependence. Most of the following discussion will apply to the range 300-1100 nm, as used for "visible" astronomy.

Charge-coupled devices can be made as front side illuminated devices, but the spectral response is cut off at short wavelengths by absorption in the semitransparent polysilicon electrodes. Peak response is also normally limited to less than 50%. For astronomical use a back side illuminated CCD is the default. With careful processing and suitable antireflection coatings, spectral response can be obtained over a wide range with a peak approaching 100%. Figure 5 illustrates front side and back side illuminated CCD constructions.

A back side illuminated CCD offers high spectral response, but only if processed carefully. Devices are commonly thinned to a thickness of 10-20 ^m, and require back surface treatment to ensure that photons absorbed near the surface are collected. Treatments that have been used include ion implantation followed by laser annealing [6], ion implantation followed by furnace annealing, chemisorption charging [7], and molecular-beam-epitaxy/delta doping.

Figure 4. CCD electromagnetic spectral range. Courtesy of e2v technologies.

UV' lijhl Visible IR < 400rm 400-700nm > 700nni

ï i Î

Frontside electrodes %5yp=ca=ÏÛÏ=K!=5yiSiS=

Depleted silicon


lindepleted silicon

UV light Visible IK < 400nm 400-700nm > 700nm

UV light Visible IK < 400nm 400-700nm > 700nm

Figure 5. (top) Front side and (bottom) back side illumination.

Figure 5. (top) Front side and (bottom) back side illumination.

An important requirement is the use of a suitable antireflection coating because silicon has a high refractive index (n~4). The usual material for a single-layer A/4 coating is hafnium oxide (n~2), which allows near-perfect photon collection especially at midband wavelengths. The curves in Fig. 6 illustrate typical spectral responses for different coatings.

Figure 6. Example of CCD responses for different antireflection coating peak wavelengths.

For a device with a thickness of 40 ^m, illustrated in Fig. 6, a common starting material is 20- or 100-Q-cm resistivity silicon; after back-thinning and processing the final thickness is typically 10-16 ^m. For long-wavelength (red) use, however, the absorption length of these photons exceeds the device thickness [3] so not all light is collected. To improve red response some manufacturers offer so-called deep-depletion CCDs; with higher-resistivity silicon, e.g., 1500 Q-cm, the device may be made with a 40-^m thickness and consequent higher red response.

A possible further step is to make devices even thicker. In this case a so-called "high-rho" device can be made of very high resistivity bulk silicon, e.g., 10,000 Q-cm. These devices generate larger depletion depths within the silicon and can also be operated with larger voltages to increase it further; a substrate voltage up to 50 V may be used in some cases. Such devices have been championed by the LBNL group [8] and also manufactured by e2v technologies and Lincoln Laboratory (originally for x-ray detection). Such devices can exhibit almost 100% QE at wavelengths around 900 nm, with an eventual reduction as the silicon band gap limit is reached. Figure 7 gives examples of spectral response for different thicknesses of silicon, and measurements [3] of LBNL 280-^m-thick samples.

Figure 7. (top) Spectral response for different silicon thicknesses; (bottom) Lick/LBNL measurements. Courtesy of e2v technologies.

Another consideration is fringing. Back-thinned devices of "normal" thickness suffer from internal multipass fringes, which modulate the response at wavelengths longward of 700 nm, where the absorption depth is comparable or longer than the silicon thickness. Devices made of progressively thicker silicon suffer less from this effect [9]. The very thick silicon depth causes these devices to collect a significantly enhanced number of cosmic ray events. The devices can potentially have a poorer PSF, and so higher operating voltages are normal in order to maximize field strength and achieve good charge confinement.

3.2 Readout Noise

Noise is the second parameter that determines the signal-to-noise of recorded data, and a low value of readout noise is often considered essential for low-signal-level astronomical applications. A low noise floor is always important for use at low signal levels, often at low pixel rates. Low noise at higher pixel rates makes the devices useful at higher frame rates, thereby reducing readout time and increasing observing efficiency.

Most scientific sensors utilize a two-stage on-chip output circuit. This allows a small first-stage transistor to couple to a small output node and provides a larger second-stage follower to provide reasonable output drive capability. Figure 8 shows an example of a two-stage (e2v technologies) output circuit. Note that an (optional) external junction field-effect transistor buffer is also shown, which is mainly beneficial for driving longer cables with appreciable capacitance.

Figure 8. Two-stage CCD output circuit schematic.

It is usual to match the designs of the two stages, and one consequence is that very low noise levels need to be traded against the output drive requirements. Large-signal use and high-frequency operation, especially to capacitative loads, require large transistors with a higher noise floor. Figure 9 gives examples for different output amplifier designs.

For scientific use, CCDs are normally expected to have readout noise floors of 2-5 e- rms, although slightly higher levels are sometimes used. A recent development of avalanche gain technology (electron multiplication) within CCDs has made subelectron readout noise available. Figure 10 illustrates an example of such a device; applications are discussed in [10].

Figure 9. Readout noise vs frequency for different responsivity amplifiers. Courtesy of e2v technologies.
Figure 10. Example of avalanche multiplication structure. Courtesy of e2v technologies.

These devices have image areas and an initial serial register that are similar to normal CCDs. However, they utilize an extended serial register with multiple stages (typically ~500), each of which allows a small probability of electron multiplication when operated with a high-voltage (up to 50 V) clock phase. The result is that the signal at the gain register may be amplified by a factor of 1000 or so before it feeds into the (standard) output stage. Thus, even an output amplifier with a high noise level can yield an input-referred readout noise in the subelectron region.

While these devices offer substantially reduced readout noise levels, appreciation of several factors is important for their full use. These include cooling to suppress dark current, control of operating temperature and highvoltage clock level for gain stability, different noise statistics resulting from the stochastic gain process, clock-induced charge at subelectron signal levels, and noise statistics different from the familiar (Gaussian) ones of traditional CCDs.

3.3 Dark Current

Charge-coupled devices collect dark current, which scales strongly with temperature. It has two main components—surface dark current and bulk dark current. The specific magnitude scales with pixel area and can depend on manufacturing process. Bulk dark current is typically 100 times lower than surface dark current. When multi-phase pinned (MPP) or inverted-mode operation (IMO) is used, devices should achieve bulk dark current levels, as seen in Fig. 11.

Figure 11. Curves showing (upper) surface and (lower) bulk dark current. Courtesy of e2v technologies.

When the device goes out of inversion, i.e., the clocks are raised above the inversion level, there is a characteristic time before the dark current recovers from bulk to surface levels. This time constant can be appreciable for low temperatures. This means that dark current can be influenced by clocking dynamics. Surface dark current can be reduced by active clocking during integration, i.e., "dither" clocking, and can also be low immediately after a previous readout at low temperatures [11a, 11].

3.4 Modulation Transfer Function

The primary function of the image sensor is to produce an output that faithfully represents the scene being imaged. The sensor must accurately reproduce all the details in the image, which contains features of varying intensities and spatial frequencies. The resolving power of the sensor is determined by its modulation transfer function (MTF), defined as the response of the sensor to a sinusoidal signal of increasing frequency. The MTF describes the decrease in contrast in the reproduced image as the spatial frequency in the original scene increases. The reduction in modulation (contrast) of closely spaced line pairs results in image blur since the separation between the light and dark lines can no longer be observed. Different methods are used to measure MTF [12]. One is to directly measure the sensor response to a sinusoidal input source. Another is to perform the Fourier transform of the PSF, which yields the optical transfer function (OTF). The MTF is the magnitude of the real part of the complex OTF variable.

Modulation transfer function is defined as the ratio of modulation depth of the output signal to the input signal:

A __ Modulation (Output)

Modulation (Input)

Modulation = SignalMAX - SignalMiN .

SignalMAX + SignalMIN

The overall MTF of the instrument is the product of the MTF of each optical component of the system, including the lens, the sensor, the electronics, and the display; however, the MTF of the sensor is usually the limiting factor.

Since the sensor is essentially a spatial sampling device, the highest frequency that it can accurately reproduce is defined by the Nyquist frequency, fNyquist = 1/2p where p is the pixel pitch. The finite sampling nature of the sensor is characterized by the sampling MTF, which is directly influenced by the pixel geometry. The MTF of the CCD is further limited by the charge-transfer inefficiency and carrier diffusion.

The overall MTF of a CCD is the product of three components: pixel geometry, charge-transfer inefficiency, and carrier lateral diffusion:

MTFccd = MTFS x MTFt x MTFd.

The discrete spacing of the pixels in the CCD places a fundamental limitation on its performance. The CCD samples the image in spatially discrete steps, and the spatial MTF is given by

MTFS = sin(nfsp)/(nfsp) = sinc(fsp), where fs is the spatial frequency and p is the pixel pitch. At the Nyquist frequency, MTFs is limited to 0.637.

Imperfect charge-transfer efficiency (CTE) results in a reduction in the output signal, causing a loss in the response amplitude, and

MTFt = exp{-ne(1 - cos 2rcf/fc)}, where f is the spatial frequency of the signal transferred through the device at the fc clock frequency, e is the charge-transfer inefficiency per transfer, and n is the number of transfers in the CCD. Even with a large number of transfers, the MTFt is usually a minor component in CCDs with high CTE.

Charge carriers that are generated inside the depleted region of the channel are driven by its electric field to the potential wells of the pixels directly above the location where they originated, but those charge carriers that are generated outside of the depleted region, in the field-free region, will diffuse randomly in the substrate and carry a high probability of being collected in the neighboring pixels instead. This effect represents the largest component of MTF degradation in the sensor. Since the photon absorption depth increases with wavelength, MTFd is worse at longer wavelengths in a front-illuminated sensor, but in a back-illuminated sensor, the spatial resolution degradation occurs at the short wavelengths.

3.5 Point Spread Function

In a well-behaved optical system, a point source of light at the object plane will generate a corresponding spot image in the image plane. The shape of the image formed at the sensor by the point source, the PSF, ideally will have a circular shape and cover only a small region. The size of the PSF is due to spreading of charge carriers, by random diffusion in the field-free region below the pixels. The width of the PSF limits the spatial resolution of the sensor, and a common practice is to report the value at full width at halfmaximum (FWHM), which is the diameter of the PSF where the signal intensity at the center of the image is reduced by half, as seen in Fig. 12.

Charge diffusion in the field-free region results in enlargement of the PSF, as shown in Fig. 13. To narrow the PSF the undesirable effects of charge diffusion must be controlled, either by reducing the size of the field-free region or by widening the depletion depth. The field-free region can be minimized by reducing the thickness of the substrate, but this is an undesirable option since the QE at long wavelengths would suffer. A better approach is to fabricate the devices on high-resistivity material, which helps extend the depth of the depletion region.

Figure 12. Diameter of PSF where signal intensity at the center of the image is reduced by half.

Frofflude of defector

Figure 13. Enlargement of PSF resulting from charge diffusion in the field-free region.

Frofflude of defector

Figure 13. Enlargement of PSF resulting from charge diffusion in the field-free region.

3.6 Deep Depletion and Fully Depleted CCDs

To improve the QE response in the red and near infrared, deep-depletion CCDs are fabricated on high-resistivity silicon material, which helps to extend the depth of the depletion layer so that long-wavelength photons are effectively absorbed. Commercially available devices are produced on 50-^m-thick, high-resistivity material, achieving a depletion depth of about

30 |m under normal operating voltages. Both front-illuminated and back-illuminated devices are offered.

Fully depleted CCDs are fabricated on 200-300-|m-thick silicon material of very high resistivity (10-12 kQ-cm). An independent bias is applied to the substrate to fully deplete the devices. The devices are back illuminated and yield exceptional QE in the near infrared, thanks to the thick substrate [3]. With the proper antireflection coating the QE at 1000 nm is about 60%, compared to a maximum of about 16% with a deep-depletion CCD.

The application of the substrate bias purges the mobile majority carriers from the substrate and generates an electric field that extends from the channel to the back side of the device. The electric field pushes the photogenerated carriers to the proper potential wells, inhibiting the lateral charge diffusion that lowers the MTF. The PSF of a 300-|m-thick fully depleted CCD at 400 nm measures 8-10 |m with 40-V substrate bias, and 6 |m when the substrate bias is increased to 77 V [13].

3.7 Radiation Tolerance

A CCD is normally capable of transferring charge with practically no loss after a very large number of transfers. This requires that the signal paths in the CCD be completely free of charge traps or other defects. When the devices operate in space, continued exposure to energetic particle radiation leads to degraded device performance. The main defect mechanisms are displacement damage and total ionizing dose effects [14]. In space applications, displacement damage effects have a stronger impact on CCD performance.

Radiation damage adds new energy levels in the band gap, facilitating the transition of electrons to the conduction band as seen in Fig. 14, and increases dark current. These defects can also trap charge and release them after some time constant, degrading the CTE. High-energy protons, neutrons, and electrons produce displacement damage when they collide with the silicon atoms, resulting in atomic defects, such as dislocations, in the crystal lattice, which cause an increase in dark current, generate hot pixels, and lower CTE because of the additional charge-trapping sites.

Conduction band

Conduction band

Valence band

Figure 14. Illustration of effects of radiation damage, which facilitates transition of electrons to the conduction band.

Valence band

Figure 14. Illustration of effects of radiation damage, which facilitates transition of electrons to the conduction band.

Ionizing radiation damage results in the buildup of excess positive charge in the gate dielectric, which offsets the flatband voltage, effectively changing the effect of the applied bias and clock voltages. The generation of traps at the SiO2 interface also results in an increase of dark current and a degradation of CTE.

A number of techniques have been developed to mitigate the effects of radiation damage, shown in Fig. 15, and accurate models have been developed to help predict the performance of CCDs following irradiation. Operating the CCD in inverted mode, so that the CCD surface is accumulated with holes, suppresses dark current and improves CTE, since the surface traps are filled and can no longer interact with the signal charge. This technique effectively improves the device resistance to total ionizing dose. The operating temperature and the clock frequencies can also influence the impact of radiation damage, since they affect the time constant of the charge traps and reduce their capture duty cycle. Adding a narrow notch in the CCD channel helps reduce the interaction between the signal charge packets and the trapping sites in the silicon. Introducing a sacrificial "fat zero" charge to all of the pixels in the CCD is another technique used to fill the traps and make them inactive, but this method increases the noise level.

Figure 15. Effects and defect locations from radiation damage due to ionizing and displacement damage. Courtesy of Niels Bassler, University of Aarhus.

Tests have shown that p-channel CCDs are more radiation tolerant than conventional n-channel devices. Proton irradiation tests performed on n-channel CCDs reveal the presence of traps with energy levels at 0.14, 0.23 and 0.41 eV below the conduction band. The 0.14-eV traps are due to A-centers (oxygen-vacancy complex), the 0.23-eV traps are caused by divacancies, and the 0.41-eV traps are believed to be caused by phosphorus vacancy (P-V) defects. Studies indicate that the P-V defects are responsible for the majority of the traps that degrade CTE performance in n-channel devices since their energy level is near the mid-gap level. These traps become much less effective when the device is cooled down below 180 K. The divacancies are considered to be the main defects that cause increase in bulk dark current. P-channel CCDs are not susceptible to the P-V defects and have demonstrated much greater radiation tolerance than conventional n-channel devices.

3.8 Packaging and Mosaics

Making CCDs is often easy compared to providing suitable packaging for specialized applications like astronomy. For commercial applications a simple ceramic (DIL) package can suffice and provide a cost-effective solution. For applications requiring good device flatness, metal packages are common with various connection schemes available. Large focal planes with high fill factors require the most compact buttable packages, in which a minimum footprint connector is also required. Six typical packages are shown in Fig. 16. Two large multichip applications are shown in Fig. 17.

Figure 16. (a) Ceramic package (ccd47-20), (b) Kovar package (SITe), (c) sealed Peltier package, (d) three-side buttable metal package, (e) four-side buttable + flexis, (f) custom space package. (a),(c),(d),(e),(f) Courtesy of e2v technologies.

Figure 17. (a) Vertex vxd3 detector and (b) CFHT Megacam (377 megapix).

Figure 17. (a) Vertex vxd3 detector and (b) CFHT Megacam (377 megapix).

Many applications, especially single-chip, employ simpler packages. Development costs often preclude the use in ground-based astronomy of custom packages such as those utilized in space applications. For large focal planes or "extremely large" telescopes, however, more sophisticated packages become viable. Currently, several observatories have been constructing "megacam" arrays [15], and larger ones are planned such as for the Large-Synoptic Survey Telescope (LSST) Observatory. The routine manufacture of large-area sensors in quantity has made large mosaics feasible for astronomy, which has seen a considerable growth of these within the last decade or so, as illustrated in Fig. 18.

Figure 18. Illustration of large focal-plane sizes with Luppino/Burke "Moore's" law.

Along with the growth of mosaics using "standard" CCD packages, custom packages driven by the requirements of large mosaics have also been developed, as seen in Fig. 19.

Figure 19. Examples of custom packages: (top) Luppino concept package; (middle) Lesser application-specific integrated circuit concept for LSST [16]; (bottom) LSST concept package.

3.9 Blooming Control

Because of the enormous range of object photon flux encountered in astronomy, the problem of pixel saturation, or blooming, inevitably arises in direct sky imaging. In a typical scientific CCD the well saturation causes charge to overflow into adjacent pixels above and below the illuminated pixels and to a much lesser extent across the channel stops into adjacent columns, as illustrated in Fig. 20 (left). The reason for this is that the channel stops are at the substrate or ground potential while the CCD channel, even with its gates at the most negative potential, is always a volt or so positive. Thus, there is always a path for electrons into adjacent pixels within a column that is energetically more favorable than across the channel stops. However, when the illumination is so intense that photoelectron current cannot completely drain along this path, the excess electrons will diffuse into the substrate below and be collected by pixels in adjacent columns.

Figure 20. (left) Blooming in a three-phase CCD showing charge overflowing into adjacent pixels within the column, and (right) pixel with blooming drains embedded within the channel stops to absorb pixel overload.

A method of blooming suppression involving dithering of the clock waveforms, called clocked antiblooming (CAB), was described in 1983 [17] and can be applied to any buried-channel CCD, excluding those that have a built-in charge-transfer directionality, i.e., two-phase CCDs. The technique relies on a phenomenon called charge pumping, in which excess photoelectrons are made to recombine with holes via surface states before the blooming occurs. Charge packets must be transferred back and forth within the pixel, and each transfer results in the recombination, and therefore removal from the charge packet, of electrons equal in number to the surface states at the silicon/oxide interface.

To keep up with the build-up of excess charge, the integrated charge packet must be shifted repeatedly during image integration. The question that the required cycle rate raises is one of the limitations of this method, namely, that it depends on the surface state density of the device technology. Typical values of contemporary technology lie in the low 109 cm-2, hence of order 5000 states in a 15x15-^m pixel. Thus, for a device with a 200,000 e-full well, each forward/backward shift can eliminate only 2.5 % of the pixel overload. At a 20-kHz back-and-forth shift rate this method can sink about 10 pA of bloomed charge per pixel. Of course, higher surface state densities would clearly be more attractive for this process, but manufacturers take great pains to reduce these states because they are a prime contributor to dark current.

A more attractive approach for the user is the kind of built-in blooming control first developed at what used to be RCA Laboratories (now Sarnoff Corporation) in the 1970s [18]. This feature is available from almost all the manufacturers of scientific CCDs, including e2v technologies, Fairchild Imaging, and Semiconductor Technology Associates. Figures 21(a) and 21(b) illustrate the pixel modifications used in this method. The standard p+ channel stop is modified by placing an n+ overflow or blooming drain flanked on both sides by n- regions, whose doping level is carefully controlled to produce a potential barrier or charge "spillway" that allows charge to flow from the well into the blooming drain at some predetermined level. A p+ photoelectron barrier is placed beneath these regions to deflect into the wells those photoelectrons that would otherwise be captured by the blooming drain. Figure 21(b) depicts electron potentials and charge flow when the blooming control is active. Clearly, the operation of this feature depends on the low level of the clock phases adjacent to the integrating phase in order to establish a blocking potential.

Potential under adjacent gates

Potential under adjacent gates

Figure 21. (a) Cross-sectional view of a channel stop modified for blooming control and (b) depiction of the potential profiles in the blooming-control structure.

The pixel modifications are not difficult from a process point of view, nor are they regarded as having a significant impact on device yield. The blooming control depicted here does occupy greater width than a standard channel stop and therefore results in some penalty to the full well. While standard channel stops have widths 1.5-3 ^m (after lateral diffusion of the p-dopants and the bird's beak encroachment), the blooming control as fabricated at Lincoln Laboratory [19] occupies about 4 |m. Data at Lincoln Laboratory on a 2Kx4K, 15-|m-pixel device has shown that the blooming control can handle currents of at least 1 nA from a pixel before blooming sets in, or about 100x higher than charge pumping.

Was this article helpful?

0 0
Telescopes Mastery

Telescopes Mastery

Through this ebook, you are going to learn what you will need to know all about the telescopes that can provide a fun and rewarding hobby for you and your family!

Get My Free Ebook

Post a comment